Two components are failing 10 times more than they should in the subsea space, says Sture Angelsen, business development manager for certification and risk advisory firm DNV GL; mooring lines and flexible risers and umbilicals, particularly flexible lines under higher pressure.
To help understand the issues around both, DNV GL is considering something it has being using in the maritime shipping segment called a digital twin, a model that measures the movement of the rig or semisubmersible, etc., plus all manner of other data, including the weather and operating conditions (pressure, temperature, etc.), then models the stress and strain in the equipment.
“That’s one way to see the condition prior to a failure and then look for any sign of that in the future,” Angelsen says.
DNV GL has also looked into wellhead fatigue analysis, resulting in recommended practice 0142. Part of the work here was about increasing the amount of monitoring, to provide data which could reduce uncertainty in the modeling. “This could enable better timing of recertification,” Angelsen says, as well as reduce the number of failures, which appeared to be higher once a system had been removed and opened up for recertification.
The latest subsea processing technologies are coming with monitoring technologies included, including Statoil’s Asgard subsea compression station, offshore Norway. But, Angelsen asks, can we monitor corrosion inside the Xmas tree? “There are some corrosion probes, but the technology isn’t that widely used,” he says.
It’s even harder for subsea trees, etc., installed 20+ years ago that don’t have today’s monitoring systems and are increasingly suffering failures. With high retrieval costs, this could mean loss of safety barriers and or early shut-in of wells. System obsolescence is also an issue, Angelsen says.
Ultimately, monitoring, data analysis and modeling go hand in hand. Monitoring is fine, but you need to make sure what you’re monitoring has value and that the data is accurate, Angelsen says. Then, you need staff with operational understanding who can interpret the data and if you’re using machine learning failures need to be fed into the system. “Data in itself isn’t necessarily the Holy Grail,” he says. “Machine learning only works if you can feed a failure in to it.”
It’s also important to focus on implementation, he says. In a trial looking at reliability, condition monitoring equipment was installed on a floating production unit. Some years later, the operating staff were asked if they had used the information. They hadn’t, for a number of reasons, including trust, bad information, the sensors not being properly installed, etc.